Worker Behavior Recognition Using Skeleton Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing behavior recognition systems fail to accurately identify and recognize the behavior of a specific worker when multiple workers are present in the video data, as they do not account for the distinction between different individuals.
Innovation Solution
A method that involves obtaining time series skeleton information from video data, calculating position coordinates of each worker's body parts, and identifying the worker subject to behavior recognition based on worker speed and cumulative orientation time, allowing for the exclusion of workers who are not the focus of recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If skeleton information of multiple workers is obtained from video data, then the system can capture all workers in the scene, but the behavior recognition accuracy deteriorates because the system cannot distinguish which worker is the intended subject
Solution Approach 1:
The patent segments the multiple workers detected in the video by creating individual candidate behavior recognition results for each worker. The behavior recognition unit generates separate skeleton information and behavior analysis for each detected worker, allowing the system to process multiple workers simultaneously while maintaining the ability to evaluate and select the intended subject based on additional criteria such as work state and position coordinates
Solution Approach 2:
The patent introduces new parameters to distinguish the intended worker from others: work state parameters (determining whether the worker is performing work actions) and position coordinates (comparing detected positions with predetermined work area positions). By changing the evaluation parameters from just skeleton information to including work state and position matching, the system can accurately identify the intended worker among multiple candidates
2Ease of operation
If the system recognizes behavior of any worker without distinction, then processing is simplified, but the intended worker's behavior cannot be accurately identified when multiple workers are present
Solution Approach 1:
The patent implements a feedback mechanism where the behavior recognition unit generates candidate results for each worker, then the work state determination unit and position coordinate comparison unit evaluate these candidates. The system uses feedback from work state analysis (whether the worker is performing work actions) and position matching (whether the worker is in the expected work area) to iteratively identify and confirm the intended worker, balancing simplicity with accuracy
Data Source
AI summary
In a method for recognizing behavior, a computer obtains skeleton information of a worker in time series from video data. The computer obtains position coordinates of each part of the worker included in the obtained skeleton information. When the skeleton information of more than one worker is obtained, the computer identifies a worker subject to behavior recognition based on the position coordinates of each worker. The computer recognizes behavior of the identified worker based on the time series of the skeleton information of the identified worker. In the behavior recognition method, the behavior of a worker may be recognized by using a learned model learned in advance by machine learning.